Thermal Response

[ronv]

Here is the problem:

Lets say we have a power mos fet (TO220) That runs saturated (hard on) most of the time with a 3 amp load so the DC temperature rise is only about 6C. But say evey 45 minutes it runs in linear mode for 2 seconds with a power disipation of 70 watts. How do you calculate the temperature of the junction at the end of that 2 seconds.

1- without a heatsink

2- with a heatsink if one is needed.

 

[duffy]

Ambient temperature plus the dissipated power times the thermal coefficient of the junction. In this case the ambient temperature will be the steady-state temperature of the saturated drive condition, and the dissipated power will have to be calculated from the linear mode power averaged over time. If you can use an empirical approach for this one, I would recommend doing that instead - there's lots of little fiddly crap with extra power from things like the integrated body-drain diode and parasitic bipolar effects that can throw off the calculations.

 

[ronv]

That's what I was thinking as well. The thing that is missing is a thermal response curve for the package or heatsink like they post for the junction.

 

[duffy]

Right - you need that curve or a thermal coefficient for package-to-heatsink, which is a crapshoot, and heatsink-to-ambient, which is affected by the enclosure.

 

[MrAl]

Hi,

What we could use is the specific heat capacity of the silicon die and the package metal. The rate of temperature rise depends on that, but unfortunately it also depends on the mass. I use the word 'unfortunately' here because it is unfortunate that the mass of the package and die is pretty darn small. So it would be like charging a capacitor with a small value from a voltage source through a small resistor. Thus, with a full 70 watts i think there is a good possibility it would burn to a crisp even in 2 seconds. With a heatsink of 70 square inches with perfect conduction everywhere it would heat up to close to 70 degrees C over ambient but that's only if the MOSFET was rated for that high power or better.

Doing some measurements would help. Such as pumping it with 70 watts with a heatsink and measuring some temperatures over time.

 

[duffy]

The specific heat of the silicon (0.71 J/g°K) usually isn't usually factored in, because the wafer is extremely thin and continuously bonded over a relatively large surface - so for modelling you just treat it as a boundary condition instead of using the classic volumetric approach. That way it reduces to issues with the thermal coefficient of the boundary itself, which is generally pretty low. Same with the heatsink, which also has a large area-to-volume ratio. Then the whole thing reduces to a few linear equations.

At some point this is cheating, so what I like to do is double-check how much it can really handle by sacrificing a couple of drivers to the gods of thermodynamics.

 

[ronv]

The FET is big enough. It's the one the water pump guys are using:

https://www.electro-tech-online.com/custompdfs/2012/05/A-104.pdf

I'm a little suprised that google can't find the thermal response of the TO 220 package.

Your right, some measurements seem to be in order if they want to use a linear current limit. My experience makes me think it is only about as long as it takes to see smoke.

 

[ronv]

Think I found it. .54. So If I read it right.... 70 watts / .54 = 130C in 1 second. Needs a heat sink.

https://www.electro-tech-online.com/custompdfs/2012/06/AN-7516.pdf

 

[MrAl]

Hi,


That's close to a quick estimate i came up with:
70/0.4 in 1 second. So in 2 seconds that's 350 degrees C.
70/0.5 comes out to 280 degrees C in 2 seconds.
That's counting the metal as 1 gram of copper. I'd have to check the actual size of the package, it's been a while.
Yes i figured the die was too small to think about too, and that the contact area was a pretty good heat conductor so there was no lag in the response of the die to package metal full conduction. If there is a lag, which there might be a little, the die is even hotter.